Skip to content

Erdős Problem #311

Editorial depth: Reviewed explanatory article — evidence qualified.
Snapshot classification: open (not automatically the live result).
Subject areas: number theory, unit fractions.

Read this status correctly

The status above is reproduced from GCL's immutable external-catalogue snapshot. It is not a new claim of current mathematical resolution, a source-semantic review, an independent proof check, or GCL certification. A published proof and a GCL replay are separate events.

The question in plain language

For a positive integer \(N\), choose any collection \(A\) of denominators from \(1\) through \(N\), and add their reciprocals. Problem 311 asks how closely such a sum can approach \(1\) without being exactly \(1\). Define \(delta(N)\) to be the smallest positive value of

\[ \left|1-\sum_{n\in A}\frac{1}{n}\right| \]

as \(A\) ranges over subsets of \(\{1,\ldots,N\}\). The empty set is allowed, but a zero difference is omitted. The proposed asymptotic question is whether there is a constant \(c\) with \(0<c<1\) for which

\[ \delta(N)=e^{-(c+o(1))N} \]

as \(N\) tends to infinity through the positive integers. Equivalently, the quantity \(-\log(\delta(N))/N\) should converge to \(c\).

The formulation cited from Erdős and Graham has an extra admissibility condition: \(A\) must not contain a subset \(S\) whose reciprocal sum is exactly \(1\). The live problem page attributes the observation that this is equivalent to the simpler formulation above to Vjekoslav Kovac. Intuitively, a set containing such an \(S\) has a remaining positive reciprocal contribution, at least \(1/N\); the exact equality of the two finite minima is reported on the page and in the formal file, but is not a new GCL-certified result.

A small example

The case \(N=3\) can be checked without rounding. Enumerating all eight subsets gives the following non-zero differences:

\[ \begin{array}{c|c|c} A & \sum_{n\in A}1/n & |1-\sum_{n\in A}1/n|\\ \varnothing&0&1\\ \{2\}&1/2&1/2\\ \{1,2\}&3/2&1/2\\ \{3\}&1/3&2/3\\ \{1,3\}&4/3&1/3\\ \{2,3\}&5/6&1/6\\ \{1,2,3\}&11/6&5/6 \end{array} \]

The subset \(\{1\}\) gives the excluded zero. Thus exact comparison of the remaining fractions yields \(delta(3)=1/6\). This is a reproducible finite check, not evidence that the limiting constant exists, nor a computation of its value. It also illustrates why the problem is delicate: many different subsets can have sums near \(1\), while exact equality must be separated from a very small non-zero gap.

What is known

There is an immediate lower bound. Every difference is a non-zero rational whose denominator divides \(L_N=\mathrm{lcm}(1,\ldots,N)\), so

\[ \delta(N)\geq\frac{1}{L_N}. \]

The prime number theorem gives \(\log L_N=(1+o(1))N\), hence this says only that \(delta(N)\) is at least \(e^{-(1+o(1))N}\). It does not provide the conjectured lower bound with a constant strictly below \(1\).

A primary source document for a substantial upper bound is Quanyu Tang’s A Note on Problem #311, version 2, dated 15 January 2026. Its Theorem 4.1 on PDF page 7 states that there are absolute constants \(c_0>0\) and \(N_0\) such that, for every \(N\geq N_0\),

\[ \delta(N)\leq\exp\left(-c_0\frac{N}{(\log N)^3(\log\log N)^3}\right). \]

Tang constructs an admissible set whose reciprocal sum is \(1-1/t\), with \(t\) a suitable least common multiple, and then lower-bounds \(t\). This is much smaller than the trivial reciprocal-LCM lower-bound scale, but it is still far weaker than an upper bound of the form \(e^{-cN}\) with fixed positive \(c\). The page also records Tang’s earlier version with \((\log\log N)^4\) and the later stronger exponent shown above.

The external comments contain further observations, including heuristic and partial lower-bound discussions, but they are not presented here as settled theorems. In particular, the live page still labels the problem open and reports no proof exposition or proof claim.

GCL's current state

The protected source packet in original issue #1581 freezes the catalogue at teorth/erdosproblems commit 536408716724a2402bdfa7c1d764bf83d64a8c89, file blob 4e60263a80af6207aca494d2e1388e44fa372889. That entry records informal status open, last update 2025-08-31, formal status unformalized, and formalized state no. The issue’s triage class is STATEMENT_CAPTURE_REQUIRED, and it says that no Problem 311 statement was present in the admitted FC-GDM-85F86371 provider snapshot.

The live entry has moved. It remains open, but its current catalogue revision records formalized state yes on 2026-10-12, and the page says it was edited 16 January 2026. The newly linked FormalConjectures file at commit 92761004452bedfc8e6ffe6091d92cd81c5aa58a has blob d02ea698155e6ea8a56d8ad82cbf1835205af835. It defines the simpler \(delta\), states the limiting conjecture, and records the elementary bounds with sorry; only the finite sanity theorem delta_one : delta 1 = 1 is proved in the file. Formalisation of a statement is not formal proof of the conjecture, and the V2 ledger—not this article—decides accepted coverage.

What remains

The central unresolved issue is the asymptotic lower side: neither the existence of a limit nor a matching \(e^{-cN}\) lower bound with \(0<c<1\) has been established by the sources read here. A bounded next task would first normalize the admissible and unrestricted definitions, then formalize a finite exact evaluator for \(delta(N)\) for a stated small range, and separately package Tang’s Theorem 4.1 with its exact hypotheses and dependency on Liu–Sawhney. Acceptance would require a source-linked proof transcript or a precisely delimited formal theorem; a finite table alone would not count. The task should stop if the original Erdős–Graham page 40 or the cited dependency cannot be inspected, rather than silently treating a transcription as primary evidence.

Sources and claim boundaries

The original problem citation is Erdős and Graham, Old and new problems and results in combinatorial number theory (1980), p. 40, bibliographic record MR 0592420. I verified the citation through the live entry and Tang’s reference list, but did not obtain an independent scan of page 40 in this timebox; that primary-source limitation remains explicit. The live statement and dated annotations are at erdosproblems.com/311. Tang’s readable source is the version-2 PDF at repository commit 3378153da80ff5db8962fc3ff2be083a7314a351, whose Git blob is d08cdc54fe8def843c5ab08394da39eddfb04557.

The finite enumeration above is an independently run exact-arithmetic check, not a Lean replay or mathematical certification. The Tang theorem is an external reported proof result tied to its source and page, not a GCL acceptance. The Lean research theorem is a statement containing sorry, not a checked proof. The original SOURCE RESULT/1 citation is authenticated original RESULT/1. No claim of publication, resolution, novelty, or award is made here.

Catalogue and evidence provenance

  • Pinned catalogue source: teorth/erdosproblems at 536408716724.
  • GCL issue: #1581 — protected issue/assignment identity; its labels and results are live, not reproduced automatically here.
  • GCL triage: STATEMENT_CAPTURE_REQUIRED.
  • Formal-statement coverage: No statement was confirmed in GCL's pinned formal-provider snapshot. Presence does not mean the statement or proof has been checked.
  • External mathematical status: consult current problem page and its dated bibliography.

Back to catalogue · Compendium introduction